面向社交网络的数字指纹技术研究

With the rapid development of multimedia and Internet, social network services (SNS), such as Facebook, Flickr, Youtube, Twitter and so on, have been popular all over the world in recent years. The multimedia data (image, video, audio, text, etc.) shared online has increased explosively. For instance, Flickr has collected over 5 billion images and the number is increasing by 3 million per day. Another visual content sharing website YouTube receives more than 48 hours of uploaded videos per minute. SNS bring people a lot of convenience to share their photos and video clips with others. However, they also bring ease to unauthorized use of multimedia content, such as illegal duplication, processing and redistribution, which leads to the copyright infringement and privacy leakage of users. Though the digital fingerprinting, an emerging technology, has proven to be very successful to protect the unauthorized distribution of multimedia contents, it is difficult to be directly employed in social networks. Because the digital fingerprinting has to face new challenges on the fingerprint coding, embedding and tracing when used in social network. Firstly, existing fingerprint codes belong to blind coding which is unrelated to the structure of social network. Furthermore, traditional fingerprint embedding methods are very difficult to resist both geometrical and collusion attacks simultaneously while keeping low complexity. Lastly, correlation-based fingerprint tracing algorithms are inefficient and time consuming which is unscalable to the large-scale SNS. Targeting at these challenges, we intend to conduct theory and key technology research on the following three aspects: data-aware fingerprint coding algorithm, robust and secure fingerprint embedding algorithm with low complexity, and efficient fingerprint indexing and tracing algorithm. Combining data-aware hashing with community detection, we firstly propose the community-aware fingerprint coding method. Then, by utilizing the reversible properties of linear orthogonal transformation and incorporating segmental spread spectrum modulation and geometrically invariant domain construction technologies, we propose a robust, secure and efficient fingerprint embedding approach. After that, we design a fast fingerprint detection and tracing method based on the hash table and inverted indexing. In the end, we establish a social-network-oriented digital fingerprinting application framework. The research and implementation of this project will promote the application of digital fingerprinting technology in social network, advance the interdisciplinary studies of social network and digital fingerprinting, and enrich the related theories and techniques on digital fingerprinting.

社交网络由于用户众多、传播速度快，引发大量的分享媒体被非法复制和传播，造成用户版权侵犯和隐私泄露问题。而现有的数字指纹技术并不能解决这个问题，因为社交网络服务由于其自身的特性，给数字指纹编码、嵌入和检测追踪带来新的挑战性难题。本课题将针对这些难题，从面向社交网络的数字指纹编码、嵌入、索引与快速追踪、应用框架四个方面开展相关理论及关键技术研究。拟结合高维数据感知哈希和社区检测相关理论与方法，基于图谱理论和多核学习，研究具有社区感知的组合数字指纹编码；将分段扩频调制与几何不变域构建相结合，充分利用线性正交变换的可逆特性，研究低复杂度的安全鲁棒数字指纹嵌入算法；利用哈希表和倒排索引结构，实现对数字指纹的快速检测与追踪；最后研究和建立面向社交网络的数字指纹应用系统框架。本课题的研究与实现，不仅能推动数字指纹在社交网络中的应用，还能促进社交网络与数字指纹的交叉研究，将丰富数字指纹相关的理论和技术。

社交网络由于用户众多、传播速度快，引发大量的分享媒体被非法复制和传播，造成用户版权侵犯和隐私泄露问题。而现有的数字指纹技术并不能解决这个问题，因为社交网络服务由于其自身的特性，给数字指纹编码、嵌入和检测追踪带来新的挑战性难题。本课题将针对这些难题，从面向社交网络的数字指纹编码、嵌入、索引与快速追踪、应用框架四个方面开展相关理论及关键技术研究。.课题在面向社交网络的数字指纹编码技术、数据感知哈希方法、低复杂性的安全数字指纹嵌入算法、支持海量用户的数字指纹索引构建和快速追踪技术等取得了进展。在IEEE Transactions、ACM transactions、ACM MM、ICME、IWDW等国际重要杂志和会议上发表高水平20余篇，其中SCI 期刊论文13篇、高水平国际会议论文5 篇。申报了6 项国家发明专利，已获授权5项。培养的研究生包括8名博士生和30名硕士生，其中，4名博士生和23名硕士生于项目执行期内毕业。.在面向社交网络的数字指纹编码研究中，采用数据感知哈希对社区编码，使得指纹系统可以支持扩展到上亿用户，具有良好的抗共谋攻击能力，具有较大创新，对数字指纹技术的发展具有重要贡献。此外、在利用机器学习、深度学习研究数据感知哈希方面取得了一系列成果；探索社交网络环境下分享媒体的非法拷贝的发现、追踪、认定和监视机制，构建了面向社交网络的数字指纹应用系统。这些技术和系统在企业合作、国家有关部门的战略需求方面已经得到了较好的应用与推广。.综上所述：本课题的研究推动数字指纹在社交网络中的应用研究，首次提出利用社区感知哈希来定义社区编码，有效解决抗共谋攻击难题，具有较大创新；其次利用机器学习、深度学习研究多个数据感知哈希方法，为相关交叉研究起促进作用。后续研究可以利用深度对抗神经网络对数字指纹技术的安全性和鲁棒性进一步提高。